About

I am a 4th-year PhD candidate in Language Information Processing at Yonsei University, working at the intersection of computational linguistics and human-AI interaction. My research evaluates cultural bias in LLMs and AI agents, with a focus on how Taiwanese contexts are represented and how models behave socio-pragmatically across Taiwan Mandarin, Korean, and Japanese.

My current work concerns a problem that lies beyond knowledge: how an agent's responses trigger distinct cultural scripts across users[1]. Language-specific models do mitigate Western bias, but they tend to erase the distinctions between cultures that hold different values and norms[2]. Safety alignment settles whether a model should act, not how. A model can already judge whether to act in a given context, but what I work on is whose norms it follows when it does[3][4].

If any of this overlaps with what you work on, please feel free to reach out!